i want it now!: advances in mri acquisition, reconstruction and...

©2015 MFMER | slide-1

I Want It Now!: Advances in MRI Acquisition,

Reconstruction and the Use of Priors to Enable

Fast Anatomic and Physiologic Imaging to Inform

Guidance and Adaptation Decisions

Yanle Hu

07/15/2015

©2015 MFMER | slide-2

Conflict of interest: None

©2015 MFMER | slide-3

MR-guided radiation therapy

• Provide real-time MR images during radiation therapy

©2015 MFMER | slide-4

On-board MRI unit

• Desired imaging capability

• Good temporal resolution for tracking and gating

• Good spatial resolution for contouring

• Good SNR and/or CNR

• Functional capabilities for tumor response assessment

©2015 MFMER | slide-5

MRI image acquisition

Where

©2015 MFMER | slide-6

MRI image acquisition

Where

©2015 MFMER | slide-7

Balance between temporal and spatial resolutions

• A tradeoff to be made for all MR-IGRT systems

• Tracking and gating require good temporal resolution.

• Good temporal resolution -> Less time per frame.

• MR data acquisition is in k-space. It depends on the range and rate of required k-space sampling.

• Less time per frame -> smaller k-space coverage

-> lower spatial resolution

©2015 MFMER | slide-8

Balance between temporal and spatial resolutions

©2015 MFMER | slide-9

Balance between temporal and spatial resolutions

©2015 MFMER | slide-10

K-space under-sampling

• Sampling a larger k-space in a given amount of time requires k-space under-sampling

• Only a portion of the k-space is sampled

• Artifacts show in the reconstructed images

• Additional algorithms are needed to remove artifacts

• Given the same amount of time, a larger k-space is effectively covered. Good temporal and spatial resolution can be achieved simultaneously.

©2015 MFMER | slide-11

Balance between temporal and spatial resolutions

Lustig et al, IEEE signal processing magnizine 2008 25(2):72-82.

©2015 MFMER | slide-12

What is the tradeoff in the acquisition of real-time MRI images for guidance of radiation delivery?

1%

9%

1%

6%

83% A. Temporal resolution – Spatial resolution

B. SNR – contrast

C. Patient setup – patient comfort

D. Image distortion – image acquisition speed

E. Patient throughput – image acquisition speed

©2015 MFMER | slide-13

What is the tradeoff in the acquisition of real-time MRI images for guidance of radiation delivery?

A. Temporal resolution – Spatial resolution

B. SNR – contrast

C. Patient setup – patient comfort

D. Image distortion – image acquisition speed

E. Patient throughput – image acquisition speed

©2015 MFMER | slide-14

Expedition of MRI acquisition

• Methods to accelerate MR image acquisition

• Partial k-space acquisition

• Parallel imaging

• Compressed sensing

©2015 MFMER | slide-15

Which techniques can be used to expedite image acquisition?

A. Parallel imaging

B. Partial k-space acquisition

C. Compressed sensing

D. A & B

E. A, B & C

A. B. C. D. E.

0% 0%

100%

0%0%

©2015 MFMER | slide-16

Which techniques can be used to expedite image acquisition?

A. Parallel imaging

B. Partial k-space acquisition

C. Compressed sensing

D. A & B

E. A, B & C

©2015 MFMER | slide-17

Fast MRI acquisition strategies

• Partial k-space acquisition

• Based on conjugate symmetry

• Spins are real in physical world

©2015 MFMER | slide-18

Fast MRI acquisition strategies

• Partial k-space acquisition

• Theoretically half of k-space needs to be acquired

• Max acceleration factor = 2

• In reality, phase errors can void real-value assumption. phase correction is needed.

Acq data (asym)

Acq data (sym)

Syn data

©2015 MFMER | slide-19

Fast MRI acquisition strategies

• Partial k-space acquisition

John Pauly, Stanford University, EE369C class notes

Full k-space recon Half k-space recon (9/16)

©2015 MFMER | slide-20

Which forms the basis of partial k-space acquisition

A. Coil sensitivity

B. Conjugate symmetry of the

Fourier Transform of a real

signal

C. Sparsity of a signal

D. A & C

E. None of the above

A. B. C. D. E.

1%

76%

6%

13%

5%

©2015 MFMER | slide-21

A. Coil sensitivity

B. Conjugate symmetry of the Fourier Transform of a real signal

C. Sparsity of a signal

D. A & C

E. None of the above

Ref: Feinberg DA, Hale JD, Watts JC, Kaufman L, Mark A. Halving MR imaging

time by conjugation: demonstration at 3.5 kG. Radiology. 1986

Nov;161(2):527-31.

Which forms the basis of partial k-space acquisition

©2015 MFMER | slide-22

Fast MRI acquisition strategies

• Parallel imaging

• Enabled by multi-channel receiver coil arrays

Deshmane et al. JMRI, 2012 36(1):55-72.

©2015 MFMER | slide-23

Fast MRI acquisition strategies

• Parallel imaging

• Signal detected by the coil element is a function of distance to the coil element

• Spins at a specific location create different signals in different coil element

• Signal variation in coil elements can be used to get rid of artifacts.

©2015 MFMER | slide-24

Fast MRI acquisition strategies

• Parallel imaging

• SENSE (image space)

SENSitivity Encoding

• GRAPPA (k-space)

Generalized Autocalibrating Partially Parallel Acquisition

©2015 MFMER | slide-25

Fast MRI acquisition strategies

• Parallel imaging – SENSE

A

B C=A+B

Full FOV Half FOV

©2015 MFMER | slide-26

Fast MRI acquisition strategies

• Parallel imaging – SENSE

Deshmane et al. JMRI, 2012 36(1):55-72.

©2015 MFMER | slide-27

Fast MRI acquisition strategies

• Parallel imaging – SENSE

Pruessmann et al. MRM, 1999 42:952-962.

R=2, direct recon R=2, SENSE recon Full k-space recon

©2015 MFMER | slide-28

Fast MRI acquisition strategies

• Parallel imaging – GRAPPA

Griswold et al. MRM, 2002 47:1202-1210.

©2015 MFMER | slide-29

Fast MRI acquisition strategies

• Parallel imaging – GRAPPA

Griswold et al. MRM, 2002 47:1202-1210.

R=2 direct recon Full k-space recon R=2 GRAPPA recon

©2015 MFMER | slide-30

Fast MRI acquisition strategies

• Parallel imaging – limitation

• Theoretical maximum acceleration factor = number of coil elements in the multi-channel coil arrays

• Methods won’t work if there are no sensitivity variation

• Additional SNR drop due to geometric factor

©2015 MFMER | slide-31

Which of the following coils enables the capability of parallel imaging?

A. Transmit/receive head coil

B. Body coil

C. Single channel surface coil

D. Multi-channel receiver coil

arrays

E. None of the above

A. B. C. D. E.

9%

1% 0%

89%

1%

©2015 MFMER | slide-32

Which of the following coils enables the capability of parallel imaging?

A. Transmit/receive head coil

B. Body coil

C. Single channel surface coil

D. Multi-channel receiver coil arrays

E. None of the above

Ref: Deshmane A, Gulani V, Griswold MA, Seiberlich N. Parallel MR imaging,

Journal of Magnetic Resonance Imaging, 2012 July;36(1):55-72.

©2015 MFMER | slide-33

Fast MRI acquisition strategies

• Compressed sensing

• Natural images can often be compressed with little or no perceptible loss of information

• Transform-based compression has been adopted in standards like JPEG and MPEG

• Most MR images are sparse in an appropriate transform domain

• Sparsity exists in not only still MR images but also dynamic MR images.

©2015 MFMER | slide-34

Fast MRI acquisition strategies

• Compressed sensing

• Examples of sparsifying transforms

• Finite difference

• Discrete Cosine transform (JPEG)

• Discrete wavelet transform (JPEG-2000)

©2015 MFMER | slide-35

Fast MRI acquisition strategies

• Compressed sensing

• Image reconstruction is to solve the constrained optimization problem

• Ψ sparsity can be traded with finite difference sparsity

©2015 MFMER | slide-36

Fast MRI acquisition strategies

• Compressed sensing

Lustig et al, MRM,

2007 58(6):1182-95.

©2015 MFMER | slide-37

Fast MRI acquisition strategies

• Compressed sensing

©2015 MFMER | slide-38

Which of the following statements is true regarding compressed sensing?

A. It requires multi-channel

phase array coil

B. It is always accompanied by

SNR drop

C. It exploits the sparsity which

is implicit in MR images

D. It is widely available for

clinical usage

E. All of the above

A. B. C. D. E.

2% 4%

31%

0%

62%

©2015 MFMER | slide-39

Which of the following statements is true regarding compressed sensing?

A. It requires multi-channel phase array coil

B. It is always accompanied by SNR drop

C. It exploits the sparsity which is implicit in MR images

D. It is widely available for clinical usage

E. All of the above

Ref: Lustig M, Donoho D, Pauly JM. Sparse MRI: the application of compressed

sensing for rapid MR imaging. Magnetic Resonance in Medicine. 2007

Dec;58(6):1182-95.

©2015 MFMER | slide-40

Among the techniques to expedite image acquisition, which one CAN NOT have a acceleration factor of above 2?

A. Parallel imaging

B. Partial k-space acquisition

C. Compressed sensing

D. A & B

E. None of them

A. B. C. D. E.

8%

70%

5%8%9%

©2015 MFMER | slide-41

Among the techniques to expedite image acquisition, which one CAN NOT have a acceleration factor of above 2?

A. Parallel imaging

B. Partial k-space acquisition

C. Compressed sensing

D. A & B

E. None of them

Ref: Feinberg DA, Hale JD, Watts JC, Kaufman L, Mark A. Halving MR imaging

time by conjugation: demonstration at 3.5 kG. Radiology. 1986

Nov;161(2):527-31.

©2015 MFMER | slide-42

Thank you